In the "application profile" review, I argue that aspects of the Singapore
Framework are valid and interesting independently of the specific, DCAM-based
notion of a Description Set Profile.
I suggest that the "syntactic" approach to record validation described
by the Singapore Framework is a valid alternative to the "semantic" approach
to describing constraints followed, for example, in OWL ontologies.
If each type of "validation" is... valid, are the two approaches
perhaps complementary? Should the technical approach to application profiles
taken in the Singapore Framework, as written, perhaps be broadened along the
lines proposed by Dan Brickley to include a more eclectic range of methods for
documenting metadata applications?
Tom
On Tue, May 17, 2011 at 04:58:06PM -0400, Thomas Baker wrote:
> In the meantime, alternative approaches to expressing structural constraints
> for the purpose of validating RDF-based metadata have been explored within the
> Dublin Core and Semantic Web communities. In 2007, for example, Alistair Miles
> proposed "Son of Dublin Core" -- an approach for encoding and validating
> "graph-based metadata" using a concrete XML syntax together with language for
> expressing application-specific syntax constraints over a metadata graph [12].
> The company Clark & Parsia has developed an "integrity constraint validator",
> Pellet, which treats OWL "as a schema or validation language for RDF data via
> auto-generated SPARQL queries that can be executed on any SPARQL-enabled RDF
> store" by interpreting OWL ontologies with closed-world assumptions in order to
> detect constraint violations [13]. Dan Brickley advocates a more pluralistic
> and eclectic approach to application profiles encompassing a wide range of
> human-readable and machine-processable methods for documenting metadata
> patterns, from simple lists of namespaces used to Web documents, metadata
> exemplars, and example queries [14].
...
> The DCAM-based approach to application profiles as outlined in the Singapore
> Framework was an attempt to elucidate the cycle of development from functional
> requirements through the development of a domain model, the description of
> entities in that domain model following defined constraints, and the generation
> of metadata record formats on their basis -- all in a form compatible with RDF.
> Even in the absence of a formal notion of Description Set Profiles, key aspects
> of the framework arguably retain their validity as general principles of good
> metadata design. For example, the "declaration" of metadata terms in
> vocabularies versus their "re-use" in application profiles remains a useful
> distinction. The idea of non-RDF formats with defined mappings to RDF remains
> important in the context of publishing Linked Data. Basing domain models and
> descriptive models on functional requirements remains a solid design principle.
> A generic vocabulary for expressing constraints remains a worthy goal, for
> example as a basis for automating the generation of metadata input forms and
> validators.
>
> More generally, the idea of expressing application-specific constraints in
> syntactic terms -- as components that can be validated in order to ensure the
> quality of instance metadata -- presents a valid and important alternative to
> the "semantic" style of constraint modeling used in OWL ontologies. OWL
> ontologies describe an ideal universe of related entities "in the world". In
> accordance with the "open world assumption", the semantic constraints of an
> ontology refine the description of that ideal world as a license for inferring
> additional knowledge on the basis of knowledge at hand.
>
> These syntactic and semantic notions of "constraints" have significantly
> different uses. For example, if an application profile says that the
> description of a book has only one subject heading -- a URI -- and a
> description with two subject headings is encountered, a validator will report
> an error in the record. If an OWL ontology says that a book has only one
> subject heading, and a description with two subject headings is encountered, an
> OWL reasoner will infer that the two subject-heading URIs identify the same
> subject [16]. The application profile approach is good for flagging data that
> is inconsistent with the structure of a metadata record as a document, while
> OWL ontologies are good for flagging logical inconsistencies with respect to an
> ideal construct.
>
> With the growing momentum of Linked Data, legacy record formats are being
> adapted for use in Linked Data -- sometimes by translating their constraints
> into OWL, and this is sometimes based on a misunderstanding of the nature of
> OWL. OWL ontologies define constraints by refining the conceptualization of a
> world, whereas application profiles merely refine the contents of a record.
> OWL ontologies, in other words, introduce semantic constraints that will be
> inherited by users of the ontology downstream, while application profiles
> merely control the consistency of data in a syntactic sense, leaving semantics
> untouched. If quality of control of metadata records at the moment of their
> creation, or at the moment of ingest, remains on ongoing concern, metadata
> designers may wish to revisit some of the Singapore Framework's ideas in future
> standardization efforts.
...
> [12] http://aliman.googlecode.com/svn/trunk/sodc/SoDC-0.2/index.html
> [13] http://clarkparsia.com/pellet/icv/
> [14] http://lists.w3.org/Archives/Public/public-lld/2010Nov/0114.html
> [15] http://wiki.dublincore.org/index.php/Glossary/DCMI_Abstract_Model
> [16] http://www.amberdown.net/2009/09/faq-using-rdfs-or-owl-as-a-schema-language-for-validating-rdf/
> [17] http://portal.acm.org/citation.cfm?id=1875719
|